Abstract Humans maintain learned motor skills over long time-scales?for days, years or even decades. However, little is known about how the brain achieves this stability. Some studies indicate that while motor skills can remain stable for years, the individual neurons controlling them may significantly change their firing properties over the course of hours. In another view, the tuning of individual neurons is as stable as the motor skill itself. The central hypothesis of this project is that the brain encodes learned behaviors on two distinct levels ? a mesoscopic level that is highly stable, and a microscopic level in which single neurons change and are influenced by the recent history of motor performance errors. In other words, the stability of a memory is rooted not in single neuron stability, but in network patterns that persist in spite of drifting activity in individual neurons. This project investigates this hypothesis by examining the neural basis of song in zebra finches. The neural circuits that underly song behavior are well defined, extensively studied, and in key respects homologous to the cortico-basal ganglia circuits that underly sensory-motor learning in mammals. For this project, the key value of the songbird is the stability of its behavior. A songbird can sing the same learned song with great precision for years providing a unique opportunity to examine how motor skills are preserved over long time-scales. Using new tools for stable recording from neurons, the project examines single neuron tuning and network patterns underlying song over time scales of days to months. To accelerate changes in the song motor program the project uses a brain-machine interface that generates brief bursts of noise during singing whenever the brain activates specific groups of neurons. Preliminary data reveals that birds can learn to reduce this interfering noise, and improve the quality of their songs by controlling the pattern of activity in the targeted neurons. Through the brain-machine interface and other experiments, significant preliminary data reveals that whereas mesoscopic dynamical patterns in premotor cortex are stable, individual neurons can drift in and out of the ensemble pattern, and adjust their activity to minimize performance errors. This project will reveal the rules of this process with cellular resolution. Insights gained from these experiments have the potential to impact human health. If single neurons drift in motor control, then knowing the rules that govern this drift will be critical to therapeutic interventions that promote recovery after injury, or create stable brainmachine interfaces for human prosthetics.